hw/xen_pt_msi.c

   1 /*
   2  * Copyright (c) 2007, Intel Corporation.
   3  *
   4  * This work is licensed under the terms of the GNU GPL, version 2.  See
   5  * the COPYING file in the top-level directory.
   6  *
   7  * Jiang Yunhong <yunhong.jiang@intel.com>
   8  *
   9  * This file implements direct PCI assignment to a HVM guest
  10  */
  11
  12 #include <sys/mman.h>
  13
  14 #include "xen_backend.h"
  15 #include "xen_pt.h"
  16 #include "apic-msidef.h"
  17
  18
  19 #define XEN_PT_AUTO_ASSIGN -1
  20
  21 /* shift count for gflags */
  22 #define XEN_PT_GFLAGS_SHIFT_DEST_ID        0
  23 #define XEN_PT_GFLAGS_SHIFT_RH             8
  24 #define XEN_PT_GFLAGS_SHIFT_DM             9
  25 #define XEN_PT_GFLAGSSHIFT_DELIV_MODE     12
  26 #define XEN_PT_GFLAGSSHIFT_TRG_MODE       15
  27
  28
  29 /*
  30  * Helpers
  31  */
  32
  33 static inline uint8_t msi_vector(uint32_t data)
  34 {
  35     return (data & MSI_DATA_VECTOR_MASK) >> MSI_DATA_VECTOR_SHIFT;
  36 }
  37
  38 static inline uint8_t msi_dest_id(uint32_t addr)
  39 {
  40     return (addr & MSI_ADDR_DEST_ID_MASK) >> MSI_ADDR_DEST_ID_SHIFT;
  41 }
  42
  43 static inline uint32_t msi_ext_dest_id(uint32_t addr_hi)
  44 {
  45     return addr_hi & 0xffffff00;
  46 }
  47
  48 static uint32_t msi_gflags(uint32_t data, uint64_t addr)
  49 {
  50     uint32_t result = 0;
  51     int rh, dm, dest_id, deliv_mode, trig_mode;
  52
  53     rh = (addr >> MSI_ADDR_REDIRECTION_SHIFT) & 0x1;
  54     dm = (addr >> MSI_ADDR_DEST_MODE_SHIFT) & 0x1;
  55     dest_id = msi_dest_id(addr);
  56     deliv_mode = (data >> MSI_DATA_DELIVERY_MODE_SHIFT) & 0x7;
  57     trig_mode = (data >> MSI_DATA_TRIGGER_SHIFT) & 0x1;
  58
  59     result = dest_id | (rh << XEN_PT_GFLAGS_SHIFT_RH)
  60         | (dm << XEN_PT_GFLAGS_SHIFT_DM)
  61         | (deliv_mode << XEN_PT_GFLAGSSHIFT_DELIV_MODE)
  62         | (trig_mode << XEN_PT_GFLAGSSHIFT_TRG_MODE);
  63
  64     return result;
  65 }
  66
  67 static inline uint64_t msi_addr64(XenPTMSI *msi)
  68 {
  69     return (uint64_t)msi->addr_hi << 32 | msi->addr_lo;
  70 }
  71
  72 static int msi_msix_enable(XenPCIPassthroughState *s,
  73                            uint32_t address,
  74                            uint16_t flag,
  75                            bool enable)
  76 {
  77     uint16_t val = 0;
  78
  79     if (!address) {
  80         return -1;
  81     }
  82
  83     xen_host_pci_get_word(&s->real_device, address, &val);
  84     if (enable) {
  85         val |= flag;
  86     } else {
  87         val &= ~flag;
  88     }
  89     xen_host_pci_set_word(&s->real_device, address, val);
  90     return 0;
  91 }
  92
  93 static int msi_msix_setup(XenPCIPassthroughState *s,
  94                           uint64_t addr,
  95                           uint32_t data,
  96                           int *ppirq,
  97                           bool is_msix,
  98                           int msix_entry,
  99                           bool is_not_mapped)
 100 {
 101     uint8_t gvec = msi_vector(data);
 102     int rc = 0;
 103
 104     assert((!is_msix && msix_entry == 0) || is_msix);
 105
 106     if (gvec == 0) {
 107         /* if gvec is 0, the guest is asking for a particular pirq that
 108          * is passed as dest_id */
 109         *ppirq = msi_ext_dest_id(addr >> 32) | msi_dest_id(addr);
 110         if (!*ppirq) {
 111             /* this probably identifies an misconfiguration of the guest,
 112              * try the emulated path */
 113             *ppirq = XEN_PT_UNASSIGNED_PIRQ;
 114         } else {
 115             XEN_PT_LOG(&s->dev, "requested pirq %d for MSI%s"
 116                        " (vec: %#x, entry: %#x)\n",
 117                        *ppirq, is_msix ? "-X" : "", gvec, msix_entry);
 118         }
 119     }
 120
 121     if (is_not_mapped) {
 122         uint64_t table_base = 0;
 123
 124         if (is_msix) {
 125             table_base = s->msix->table_base;
 126         }
 127
 128         rc = xc_physdev_map_pirq_msi(xen_xc, xen_domid, XEN_PT_AUTO_ASSIGN,
 129                                      ppirq, PCI_DEVFN(s->real_device.dev,
 130                                                       s->real_device.func),
 131                                      s->real_device.bus,
 132                                      msix_entry, table_base);
 133         if (rc) {
 134             XEN_PT_ERR(&s->dev,
 135                        "Mapping of MSI%s (rc: %i, vec: %#x, entry %#x)\n",
 136                        is_msix ? "-X" : "", rc, gvec, msix_entry);
 137             return rc;
 138         }
 139     }
 140
 141     return 0;
 142 }
 143 static int msi_msix_update(XenPCIPassthroughState *s,
 144                            uint64_t addr,
 145                            uint32_t data,
 146                            int pirq,
 147                            bool is_msix,
 148                            int msix_entry,
 149                            int *old_pirq)
 150 {
 151     PCIDevice *d = &s->dev;
 152     uint8_t gvec = msi_vector(data);
 153     uint32_t gflags = msi_gflags(data, addr);
 154     int rc = 0;
 155     uint64_t table_addr = 0;
 156
 157     XEN_PT_LOG(d, "Updating MSI%s with pirq %d gvec %#x gflags %#x"
 158                " (entry: %#x)\n",
 159                is_msix ? "-X" : "", pirq, gvec, gflags, msix_entry);
 160
 161     if (is_msix) {
 162         table_addr = s->msix->mmio_base_addr;
 163     }
 164
 165     rc = xc_domain_update_msi_irq(xen_xc, xen_domid, gvec,
 166                                   pirq, gflags, table_addr);
 167
 168     if (rc) {
 169         XEN_PT_ERR(d, "Updating of MSI%s failed. (rc: %d)\n",
 170                    is_msix ? "-X" : "", rc);
 171
 172         if (xc_physdev_unmap_pirq(xen_xc, xen_domid, *old_pirq)) {
 173             XEN_PT_ERR(d, "Unmapping of MSI%s pirq %d failed.\n",
 174                        is_msix ? "-X" : "", *old_pirq);
 175         }
 176         *old_pirq = XEN_PT_UNASSIGNED_PIRQ;
 177     }
 178     return rc;
 179 }
 180
 181 static int msi_msix_disable(XenPCIPassthroughState *s,
 182                             uint64_t addr,
 183                             uint32_t data,
 184                             int pirq,
 185                             bool is_msix,
 186                             bool is_binded)
 187 {
 188     PCIDevice *d = &s->dev;
 189     uint8_t gvec = msi_vector(data);
 190     uint32_t gflags = msi_gflags(data, addr);
 191     int rc = 0;
 192
 193     if (pirq == XEN_PT_UNASSIGNED_PIRQ) {
 194         return 0;
 195     }
 196
 197     if (is_binded) {
 198         XEN_PT_LOG(d, "Unbind MSI%s with pirq %d, gvec %#x\n",
 199                    is_msix ? "-X" : "", pirq, gvec);
 200         rc = xc_domain_unbind_msi_irq(xen_xc, xen_domid, gvec, pirq, gflags);
 201         if (rc) {
 202             XEN_PT_ERR(d, "Unbinding of MSI%s failed. (pirq: %d, gvec: %#x)\n",
 203                        is_msix ? "-X" : "", pirq, gvec);
 204             return rc;
 205         }
 206     }
 207
 208     XEN_PT_LOG(d, "Unmap MSI%s pirq %d\n", is_msix ? "-X" : "", pirq);
 209     rc = xc_physdev_unmap_pirq(xen_xc, xen_domid, pirq);
 210     if (rc) {
 211         XEN_PT_ERR(d, "Unmapping of MSI%s pirq %d failed. (rc: %i)\n",
 212                    is_msix ? "-X" : "", pirq, rc);
 213         return rc;
 214     }
 215
 216     return 0;
 217 }
 218
 219 /*
 220  * MSI virtualization functions
 221  */
 222
 223 int xen_pt_msi_set_enable(XenPCIPassthroughState *s, bool enable)
 224 {
 225     XEN_PT_LOG(&s->dev, "%s MSI.\n", enable ? "enabling" : "disabling");
 226
 227     if (!s->msi) {
 228         return -1;
 229     }
 230
 231     return msi_msix_enable(s, s->msi->ctrl_offset, PCI_MSI_FLAGS_ENABLE,
 232                            enable);
 233 }
 234
 235 /* setup physical msi, but don't enable it */
 236 int xen_pt_msi_setup(XenPCIPassthroughState *s)
 237 {
 238     int pirq = XEN_PT_UNASSIGNED_PIRQ;
 239     int rc = 0;
 240     XenPTMSI *msi = s->msi;
 241
 242     if (msi->initialized) {
 243         XEN_PT_ERR(&s->dev,
 244                    "Setup physical MSI when it has been properly initialized.\n");
 245         return -1;
 246     }
 247
 248     rc = msi_msix_setup(s, msi_addr64(msi), msi->data, &pirq, false, 0, true);
 249     if (rc) {
 250         return rc;
 251     }
 252
 253     if (pirq < 0) {
 254         XEN_PT_ERR(&s->dev, "Invalid pirq number: %d.\n", pirq);
 255         return -1;
 256     }
 257
 258     msi->pirq = pirq;
 259     XEN_PT_LOG(&s->dev, "MSI mapped with pirq %d.\n", pirq);
 260
 261     return 0;
 262 }
 263
 264 int xen_pt_msi_update(XenPCIPassthroughState *s)
 265 {
 266     XenPTMSI *msi = s->msi;
 267     return msi_msix_update(s, msi_addr64(msi), msi->data, msi->pirq,
 268                            false, 0, &msi->pirq);
 269 }
 270
 271 void xen_pt_msi_disable(XenPCIPassthroughState *s)
 272 {
 273     XenPTMSI *msi = s->msi;
 274
 275     if (!msi) {
 276         return;
 277     }
 278
 279     xen_pt_msi_set_enable(s, false);
 280
 281     msi_msix_disable(s, msi_addr64(msi), msi->data, msi->pirq, false,
 282                      msi->initialized);
 283
 284     /* clear msi info */
 285     msi->flags = 0;
 286     msi->mapped = false;
 287     msi->pirq = XEN_PT_UNASSIGNED_PIRQ;
 288 }
 289
 290 /*
 291  * MSI-X virtualization functions
 292  */
 293
 294 static int msix_set_enable(XenPCIPassthroughState *s, bool enabled)
 295 {
 296     XEN_PT_LOG(&s->dev, "%s MSI-X.\n", enabled ? "enabling" : "disabling");
 297
 298     if (!s->msix) {
 299         return -1;
 300     }
 301
 302     return msi_msix_enable(s, s->msix->ctrl_offset, PCI_MSIX_FLAGS_ENABLE,
 303                            enabled);
 304 }
 305
 306 static int xen_pt_msix_update_one(XenPCIPassthroughState *s, int entry_nr)
 307 {
 308     XenPTMSIXEntry *entry = NULL;
 309     int pirq;
 310     int rc;
 311
 312     if (entry_nr < 0 || entry_nr >= s->msix->total_entries) {
 313         return -EINVAL;
 314     }
 315
 316     entry = &s->msix->msix_entry[entry_nr];
 317
 318     if (!entry->updated) {
 319         return 0;
 320     }
 321
 322     pirq = entry->pirq;
 323
 324     rc = msi_msix_setup(s, entry->addr, entry->data, &pirq, true, entry_nr,
 325                         entry->pirq == XEN_PT_UNASSIGNED_PIRQ);
 326     if (rc) {
 327         return rc;
 328     }
 329     if (entry->pirq == XEN_PT_UNASSIGNED_PIRQ) {
 330         entry->pirq = pirq;
 331     }
 332
 333     rc = msi_msix_update(s, entry->addr, entry->data, pirq, true,
 334                          entry_nr, &entry->pirq);
 335
 336     if (!rc) {
 337         entry->updated = false;
 338     }
 339
 340     return rc;
 341 }
 342
 343 int xen_pt_msix_update(XenPCIPassthroughState *s)
 344 {
 345     XenPTMSIX *msix = s->msix;
 346     int i;
 347
 348     for (i = 0; i < msix->total_entries; i++) {
 349         xen_pt_msix_update_one(s, i);
 350     }
 351
 352     return 0;
 353 }
 354
 355 void xen_pt_msix_disable(XenPCIPassthroughState *s)
 356 {
 357     int i = 0;
 358
 359     msix_set_enable(s, false);
 360
 361     for (i = 0; i < s->msix->total_entries; i++) {
 362         XenPTMSIXEntry *entry = &s->msix->msix_entry[i];
 363
 364         msi_msix_disable(s, entry->addr, entry->data, entry->pirq, true, true);
 365
 366         /* clear MSI-X info */
 367         entry->pirq = XEN_PT_UNASSIGNED_PIRQ;
 368         entry->updated = false;
 369     }
 370 }
 371
 372 int xen_pt_msix_update_remap(XenPCIPassthroughState *s, int bar_index)
 373 {
 374     XenPTMSIXEntry *entry;
 375     int i, ret;
 376
 377     if (!(s->msix && s->msix->bar_index == bar_index)) {
 378         return 0;
 379     }
 380
 381     for (i = 0; i < s->msix->total_entries; i++) {
 382         entry = &s->msix->msix_entry[i];
 383         if (entry->pirq != XEN_PT_UNASSIGNED_PIRQ) {
 384             ret = xc_domain_unbind_pt_irq(xen_xc, xen_domid, entry->pirq,
 385                                           PT_IRQ_TYPE_MSI, 0, 0, 0, 0);
 386             if (ret) {
 387                 XEN_PT_ERR(&s->dev, "unbind MSI-X entry %d failed\n",
 388                            entry->pirq);
 389             }
 390             entry->updated = true;
 391         }
 392     }
 393     return xen_pt_msix_update(s);
 394 }
 395
 396 static uint32_t get_entry_value(XenPTMSIXEntry *e, int offset)
 397 {
 398     switch (offset) {
 399     case PCI_MSIX_ENTRY_LOWER_ADDR:
 400         return e->addr & UINT32_MAX;
 401     case PCI_MSIX_ENTRY_UPPER_ADDR:
 402         return e->addr >> 32;
 403     case PCI_MSIX_ENTRY_DATA:
 404         return e->data;
 405     case PCI_MSIX_ENTRY_VECTOR_CTRL:
 406         return e->vector_ctrl;
 407     default:
 408         return 0;
 409     }
 410 }
 411
 412 static void set_entry_value(XenPTMSIXEntry *e, int offset, uint32_t val)
 413 {
 414     switch (offset) {
 415     case PCI_MSIX_ENTRY_LOWER_ADDR:
 416         e->addr = (e->addr & ((uint64_t)UINT32_MAX << 32)) | val;
 417         break;
 418     case PCI_MSIX_ENTRY_UPPER_ADDR:
 419         e->addr = (uint64_t)val << 32 | (e->addr & UINT32_MAX);
 420         break;
 421     case PCI_MSIX_ENTRY_DATA:
 422         e->data = val;
 423         break;
 424     case PCI_MSIX_ENTRY_VECTOR_CTRL:
 425         e->vector_ctrl = val;
 426         break;
 427     }
 428 }
 429
 430 static void pci_msix_write(void *opaque, hwaddr addr,
 431                            uint64_t val, unsigned size)
 432 {
 433     XenPCIPassthroughState *s = opaque;
 434     XenPTMSIX *msix = s->msix;
 435     XenPTMSIXEntry *entry;
 436     int entry_nr, offset;
 437
 438     entry_nr = addr / PCI_MSIX_ENTRY_SIZE;
 439     if (entry_nr < 0 || entry_nr >= msix->total_entries) {
 440         XEN_PT_ERR(&s->dev, "asked MSI-X entry '%i' invalid!\n", entry_nr);
 441         return;
 442     }
 443     entry = &msix->msix_entry[entry_nr];
 444     offset = addr % PCI_MSIX_ENTRY_SIZE;
 445
 446     if (offset != PCI_MSIX_ENTRY_VECTOR_CTRL) {
 447         const volatile uint32_t *vec_ctrl;
 448
 449         if (get_entry_value(entry, offset) == val) {
 450             return;
 451         }
 452
 453         /*
 454          * If Xen intercepts the mask bit access, entry->vec_ctrl may not be
 455          * up-to-date. Read from hardware directly.
 456          */
 457         vec_ctrl = s->msix->phys_iomem_base + entry_nr * PCI_MSIX_ENTRY_SIZE
 458             + PCI_MSIX_ENTRY_VECTOR_CTRL;
 459
 460         if (msix->enabled && !(*vec_ctrl & PCI_MSIX_ENTRY_CTRL_MASKBIT)) {
 461             XEN_PT_ERR(&s->dev, "Can't update msix entry %d since MSI-X is"
 462                        " already enabled.\n", entry_nr);
 463             return;
 464         }
 465
 466         entry->updated = true;
 467     }
 468
 469     set_entry_value(entry, offset, val);
 470
 471     if (offset == PCI_MSIX_ENTRY_VECTOR_CTRL) {
 472         if (msix->enabled && !(val & PCI_MSIX_ENTRY_CTRL_MASKBIT)) {
 473             xen_pt_msix_update_one(s, entry_nr);
 474         }
 475     }
 476 }
 477
 478 static uint64_t pci_msix_read(void *opaque, hwaddr addr,
 479                               unsigned size)
 480 {
 481     XenPCIPassthroughState *s = opaque;
 482     XenPTMSIX *msix = s->msix;
 483     int entry_nr, offset;
 484
 485     entry_nr = addr / PCI_MSIX_ENTRY_SIZE;
 486     if (entry_nr < 0) {
 487         XEN_PT_ERR(&s->dev, "asked MSI-X entry '%i' invalid!\n", entry_nr);
 488         return 0;
 489     }
 490
 491     offset = addr % PCI_MSIX_ENTRY_SIZE;
 492
 493     if (addr < msix->total_entries * PCI_MSIX_ENTRY_SIZE) {
 494         return get_entry_value(&msix->msix_entry[entry_nr], offset);
 495     } else {
 496         /* Pending Bit Array (PBA) */
 497         return *(uint32_t *)(msix->phys_iomem_base + addr);
 498     }
 499 }
 500
 501 static const MemoryRegionOps pci_msix_ops = {
 502     .read = pci_msix_read,
 503     .write = pci_msix_write,
 504     .endianness = DEVICE_NATIVE_ENDIAN,
 505     .valid = {
 506         .min_access_size = 4,
 507         .max_access_size = 4,
 508         .unaligned = false,
 509     },
 510 };
 511
 512 int xen_pt_msix_init(XenPCIPassthroughState *s, uint32_t base)
 513 {
 514     uint8_t id = 0;
 515     uint16_t control = 0;
 516     uint32_t table_off = 0;
 517     int i, total_entries, bar_index;
 518     XenHostPCIDevice *hd = &s->real_device;
 519     PCIDevice *d = &s->dev;
 520     int fd = -1;
 521     XenPTMSIX *msix = NULL;
 522     int rc = 0;
 523
 524     rc = xen_host_pci_get_byte(hd, base + PCI_CAP_LIST_ID, &id);
 525     if (rc) {
 526         return rc;
 527     }
 528
 529     if (id != PCI_CAP_ID_MSIX) {
 530         XEN_PT_ERR(d, "Invalid id %#x base %#x\n", id, base);
 531         return -1;
 532     }
 533
 534     xen_host_pci_get_word(hd, base + PCI_MSIX_FLAGS, &control);
 535     total_entries = control & PCI_MSIX_FLAGS_QSIZE;
 536     total_entries += 1;
 537
 538     s->msix = g_malloc0(sizeof (XenPTMSIX)
 539                         + total_entries * sizeof (XenPTMSIXEntry));
 540     msix = s->msix;
 541
 542     msix->total_entries = total_entries;
 543     for (i = 0; i < total_entries; i++) {
 544         msix->msix_entry[i].pirq = XEN_PT_UNASSIGNED_PIRQ;
 545     }
 546
 547     memory_region_init_io(&msix->mmio, &pci_msix_ops, s, "xen-pci-pt-msix",
 548                           (total_entries * PCI_MSIX_ENTRY_SIZE
 549                            + XC_PAGE_SIZE - 1)
 550                           & XC_PAGE_MASK);
 551
 552     xen_host_pci_get_long(hd, base + PCI_MSIX_TABLE, &table_off);
 553     bar_index = msix->bar_index = table_off & PCI_MSIX_FLAGS_BIRMASK;
 554     table_off = table_off & ~PCI_MSIX_FLAGS_BIRMASK;
 555     msix->table_base = s->real_device.io_regions[bar_index].base_addr;
 556     XEN_PT_LOG(d, "get MSI-X table BAR base 0x%"PRIx64"\n", msix->table_base);
 557
 558     fd = open("/dev/mem", O_RDWR);
 559     if (fd == -1) {
 560         rc = -errno;
 561         XEN_PT_ERR(d, "Can't open /dev/mem: %s\n", strerror(errno));
 562         goto error_out;
 563     }
 564     XEN_PT_LOG(d, "table_off = %#x, total_entries = %d\n",
 565                table_off, total_entries);
 566     msix->table_offset_adjust = table_off & 0x0fff;
 567     msix->phys_iomem_base =
 568         mmap(NULL,
 569              total_entries * PCI_MSIX_ENTRY_SIZE + msix->table_offset_adjust,
 570              PROT_READ,
 571              MAP_SHARED | MAP_LOCKED,
 572              fd,
 573              msix->table_base + table_off - msix->table_offset_adjust);
 574     close(fd);
 575     if (msix->phys_iomem_base == MAP_FAILED) {
 576         rc = -errno;
 577         XEN_PT_ERR(d, "Can't map physical MSI-X table: %s\n", strerror(errno));
 578         goto error_out;
 579     }
 580     msix->phys_iomem_base = (char *)msix->phys_iomem_base
 581         + msix->table_offset_adjust;
 582
 583     XEN_PT_LOG(d, "mapping physical MSI-X table to %p\n",
 584                msix->phys_iomem_base);
 585
 586     memory_region_add_subregion_overlap(&s->bar[bar_index], table_off,
 587                                         &msix->mmio,
 588                                         2); /* Priority: pci default + 1 */
 589
 590     return 0;
 591
 592 error_out:
 593     memory_region_destroy(&msix->mmio);
 594     g_free(s->msix);
 595     s->msix = NULL;
 596     return rc;
 597 }
 598
 599 void xen_pt_msix_delete(XenPCIPassthroughState *s)
 600 {
 601     XenPTMSIX *msix = s->msix;
 602
 603     if (!msix) {
 604         return;
 605     }
 606
 607     /* unmap the MSI-X memory mapped register area */
 608     if (msix->phys_iomem_base) {
 609         XEN_PT_LOG(&s->dev, "unmapping physical MSI-X table from %p\n",
 610                    msix->phys_iomem_base);
 611         munmap(msix->phys_iomem_base, msix->total_entries * PCI_MSIX_ENTRY_SIZE
 612                + msix->table_offset_adjust);
 613     }
 614
 615     memory_region_del_subregion(&s->bar[msix->bar_index], &msix->mmio);
 616     memory_region_destroy(&msix->mmio);
 617
 618     g_free(s->msix);
 619     s->msix = NULL;
 620 }